Wiki
Clone wikignrs / Minutes 2012-11-29
Status Report
- Feixiong got 56 nodes working for an experiment.
- 40 nodes were able to run the GNRS server
Discussion
- Generating topologies
- What to evaluate
- Latency per request (at client)
- With and without cache
- Throughput vs number of requests
- Staleness (update latency)
- Consistency
- Failure
- Definition of failure
- Rate of failure
- How to handle offline nodes (TTL)
- Latency per request (at client)
- K-replica policies
- Lowest-latency replica only
- Two random, use the first response
- Cache replacement policies need to be revisited
- Simple LRU for now
- Time to live
- TTL and expire time for each binding
New Tasks
- Get "latency" values for 50 nodes on the grid
- Fully-connected, random-delay topology (5 nodes, 10 nodes, 20 nodes, 50 nodes)
- need to create an gnrs image that can work for different types of i7 node
Updated